菅原 斉

<jats:title>Abstract</jats:title> <jats:p>The applicability of existing Convolutional Neural Network (CNN) models for diagnosing heart failure in Japanese patients remains unclear. We aimed to evaluate the accuracy of a CNN-based single-lead electrocardiographic device for point-of-care heart failure screening in primary care settings. We included patients aged ≥ 20 years with a chief complaint of dyspnea or lower leg edema. A single-lead electrocardiogram was recorded, following which a CNN-based algorithm derived a quantitative index of heart failure severity (HF index). The primary endpoint was the diagnostic accuracy of the HF index. Secondary endpoints included identifying factors affecting the diagnostic performance and accuracy of the HF index in classifying heart failure phenotypes and disease severity. We enrolled 112 patients, including 50 patients with heart failure. The HF index had excellent diagnostic performance, with an area under the receiver operating characteristic curve of 0.905. Its sensitivity and specificity for heart failure diagnosis were 0.800 and 0.871, respectively. No significant interactions were observed with the HF index for several evaluated factors. The HF index exhibited strong predictive ability for heart failure phenotypes and disease severity. In older Japanese patients, the single-lead electrocardiographic device enabled simple and accessible screening for different heart failure phenotypes in primary care setting.</jats:p> <jats:p>Trial registration: This manuscript reports the results of a clinical study registered with UMIN-CTR (UMIN000057871) on May 15, 2025.</jats:p>

Background: Whether corrective actions documented in medical safety incident reports rely on individual vigilance ("Safety-I") or on structural, system-level intervention ("Safety-II") has not been quantitatively evaluated on a national scale in Japan. We developed an automated classification pipeline to assign corrective-action free-text to a 7-level maturity scale (L0-L6) and computed two summary indices: the Safety Measure Quality Profile (SMQP), the full L0-L6 distribution, and the System-based Safety Measure Rate (SSMR), the proportion of non-L0 records classified L3-L6. Methods: We analyzed all 11,507 corrective-action free-text entries from the 2010 release of Japan's national medical accident and near-miss reporting database (Japan Council for Quality Health Care, JCQHC), comprising 8,804 near-miss (Hiyari-Hatto) and 2,703 accident (Jiko) reports. Records were classified using a five-stage hybrid pipeline: an expert-developed rule dictionary, TF-IDF + k-nearest-neighbor matching, cosine-similarity matching, a two-tier large-language-model (LLM) classifier, and a conservative priority-cascade fallback. SSMR was compared between near-miss and accident reports using a chi-square test, Wilson 95% confidence intervals, Cramer's V, and the risk difference (RD), against pre-specified minimal clinically important difference (MCID) criteria of RD >= 2 percentage points and Cramer's V >= 0.10. Results: Every record received a definitive L0-L6 label (0% unresolved). Overall, 16.6% of records were unclassifiable (L0); among the 9,599 classifiable (non-L0) records, individual-vigilance actions (L1) predominated (54.6% of all records), and only 11.82% (95% CI, 11.19-12.49%) met the SSMR criterion (L3-L6). SSMR was higher for accident reports than for near-miss reports (18.12% [95% CI, 16.69-19.64%] vs. 9.[95% CI,46% [95% CI, 8.80-10.17%]; RD = 8.66 percentage points; Cramer's V = 0.120; chi-square(1) = 136.97001), exceeding both pre-specified MCID thresholds. Conclusions: In this interim single-year analysis, the large majority of documented corrective actions in Japanese medical safety reports remained individual-vigilance-based rather than system-based, with accident reports showing a substantively, rather than merely statistically, higher proportion of system-based actions than near-miss reports. These findings support the feasibility of large-scale automated assessment of corrective-action quality and provide the rationale for the planned 16-year longitudinal analysis.

ABSTRACT Background Despite the high prognostic value of D‐dimer in various clinical conditions, limited research has addressed short‐term fatality prediction across disease categories. This study aimed to develop and compare models predicting 72‐h fatality in patients with D‐dimer levels ≥ 2 μg/mL, using laboratory variables. This timeframe was chosen based on its clinical relevance for early triage and intervention across multiple acute conditions. Methods We retrospectively analyzed data from 5158 patients (241 deaths within 72 h). The primary outcome was 72‐h fatality; predictors included age, sex, and 40 routine hematologic, biochemical, and coagulation tests. Traditional multivariate logistic regression analysis (MLRA) was compared with four machine learning (ML) models: Prediction One, LightGBM, XGBoost, and CatBoost. External validation was performed using a separate dataset of 5550 patients (309 deaths). D‐dimer levels were recorded in any clinical setting despite limited patient medical information. Results The 72‐h fatality rate increased with increasing D‐dimer levels (overall 4.67%). Major causes of death were intracranial disease (24.9%), malignancy (17.0%), and sepsis (8.3%). MLRA identified five key predictors: advanced age, low total protein and cholesterol levels, and elevated aspartate aminotransferase and D‐dimer levels. Its performance (AUC 0.829, 95% CI 0.768–0.888; sensitivity 0.762; specificity 0.809) was exceeded by LightGBM (AUC 0.987; sensitivity 0.987; specificity 0.911), which outperformed Prediction One (0.814), XGBoost (0.981), and CatBoost (0.937). Conclusion ML models, particularly LightGBM, effectively identify high‐risk patients using routine laboratory tests. The model enables timely decision‐making and early risk stratification in patients with high D‐dimer values, even when clinical information is limited.